Author + information
- Rollo P. Villareal, MD (, )
- Vei-Vei Lee, MS,
- MacArthur Elayda, MD, PhD,
- James M. Wilson, MD and
- Ali Massumi, MD
We thank Dr. Habib and colleagues for their interest in our study (1) and deeply appreciate their comments. Our response is both general and specific.
In general, there will always be residual confounding unless a study population is randomized. Conversely, in randomized studies, the “confounding” takes the form of a selection bias dictated by the usually rigid inclusion and exclusion criteria. The appeal of observational database studies such as ours lies in the large population available for study, capturing the influence of real-world, more current medical practice, the availability of proven statistical methods to adjust for baseline differences in clinical characteristics, and a long history of findings utilizing these methods that predict, are consistent with, and have subsequently been corroborated by randomized trials. Indeed, the backbone of current cardiovascular risk assessment—with factors such as age, gender, diabetes, hypertension, hypercholesterolemia, tobacco use, and a family history of premature coronary disease—was derived using similar methodologies.
The comments regarding age, antiarrhythmic drug effect, and the matching process are well founded, on-the-mark, and are issues that we too have discussed at length among ourselves. We are in agreement that these may, at least in part, simply be a by-product of the methodology, which has not uncommonly been described as “statistical fishing.” Hence, the caution admonished in its interpretation. The matching process was an attempt to link as many of the patient characteristics as possible that predicted adverse outcomes, while maintaining a “respectable” number of patients for comparison, which we arbitrarily set at 200. The difficulty in matching, which involved manually entering and removing variables from the model without reducing the number significantly less than 200, probably resulted in a case-matched population with significantly different characteristics from the parent populations. This underscores the difficulty in studying this particular disease entity, but also suggests that more sensitive and specific multivariate models can be constructed to better identify patients at risk.
Overall, notwithstanding the limitations inherent in the methods and those described above, we believe we have built a strong case against postoperative atrial fibrillation as a predictor of mortality. More importantly, we have raised awareness that this is more than likely not a benign process and that it deserves more in-depth study.
- American College of Cardiology Foundation